SM

7.2K posts

SM banner
SM

SM

@pixelr_222

Like tech and occasionally touching grass 🌱

Katılım Nisan 2015
854 Takip Edilen71 Takipçiler
SM
SM@pixelr_222·
@VardyParty2025 @UnionJacked__ Your monthly takehome is 2.4k Average monthly bills (mortgage, energy etc) is 1k-1.5k excluding food and transport So if with a partner maybe explains But single person on 35k isn't going to have mortgage + food + car + much disposable
English
0
0
0
1
Bring Your Vodka And Your Charlie
@UnionJacked__ Or you're just a fucking wet wipe. I'm on 35k a year and manage to have nearly 1k disposable income each month after expenses (including mortgage) and my partner has even more. I invest, overpay the mortgage and go abroad every year. If I can do that on 35k then 50k is plenty
English
4
0
5
562
SM
SM@pixelr_222·
@putthetellyon This insulates the BBC from spend criticism. If they put it out to tender, and a commercial provider (which needs to make profit) can't do it for less then BBC Studios does it, it justifies that it's good value for the spend. If the save money, also best value of TV licence.
English
0
0
0
1
Zedmor🇺🇦
Zedmor🇺🇦@Zedmor·
@sksq96 Well…. It had… to… solve… that… puzzle… and it did so!
English
4
0
95
19.5K
Shubham
Shubham@sksq96·
so, let’s get this straight… an unreleased (gpt-6?) model was being tested in a no-internet scenario on openai’s servers. and it found several zero-day vulnerabilities in the sandbox, which was supposed to not have internet access… found internet access, went to hugging face’s servers, found another exploit in their dataset loader… just to do what exactly? game and reward-hack the ExploitGym benchmark??? so gpt-6 would basically go ahead and exploit several zero-day vulnerabilities to cheat on the exam instead of actually solving the exam
OpenAI@OpenAI

We're partnering with @huggingface to investigate an unprecedented security incident. Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation. Sharing preliminary findings to help defenders understand emerging risks: openai.com/index/hugging-…

English
178
123
4K
618.5K
SM
SM@pixelr_222·
@taziksh @GaryMarcus It means when an LLM does something I said it couldn't, I'll arbitrarily move the goalposts such that I'm still right. Or arithmetically. Heads I win, tails you lose.
English
0
0
0
2
SM
SM@pixelr_222·
@ivorytowerjourn @glosswitch Misogyny was relevant insofar as she was vulnerable. But a near 80 year old man living alone would similarly be vulnerable. Why drive 300 miles to kill specifically her? I'm sure old vulnerable ladies exist locally. Perchance it wasn't the motivation?
English
0
0
0
0
doucheen
doucheen@doucheen·
@rob_mcrobberson the hard parts can be delegated to business people now with ai. no devs needed.
English
0
0
0
55
SM
SM@pixelr_222·
@hasen_95dx @rob_mcrobberson No he's not just of his mind. And it can, but I granularly prompt with AI. I don't let it do whatever the fuck it wants. I'm the engineer. It's the code monkey.
English
0
0
0
0
ハセン حسن
ハセン حسن@hasen_95dx·
@rob_mcrobberson Dude, are you out of your mind? AI not just writes the code, it figures out what code needs to be written. 90% of the code my AI writes for me I don't know off the top of my head and would need to spend days researching to figure it out.
English
2
0
2
87
SM
SM@pixelr_222·
@solopribuilds @rob_mcrobberson Syntax isn't the toughest. Junior devs can write syntax. Not expected anywhere near as much with senior engineers.
English
0
0
0
0
SM
SM@pixelr_222·
@CKodatt69897 @attrc Isn't the Oval Office requiring these guardrails though? So it's not OAI or Anthropic who can't figure it out.
English
0
0
0
1
Charlie Kodatt
Charlie Kodatt@CKodatt69897·
Supposedly Open AI and Anthropic have hired the most brilliant people in the world, given them tens of millions of dollars a year in salary, sometimes billions, and they can't figure this out. A Chinese model has to come in and save the day. You wonder what you're paying for. You wonder if you're an investor in these companies, such as Open AI and Anthropic, shaking your head thinking, "Why in the hell did I invest in these companies?"
English
3
0
113
11.9K
Andrew Case
Andrew Case@attrc·
To summarize: HuggingFace got autonomously compromised by a model from an American company. HF then tried to use American frontier model(s) to defend themselves, but were blocked by guardrails. HF then had to turn to open source Chinese models to defend themselves from another American company.
OpenAI@OpenAI

We're partnering with @huggingface to investigate an unprecedented security incident. Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation. Sharing preliminary findings to help defenders understand emerging risks: openai.com/index/hugging-…

English
187
1.6K
15.7K
1M
SM
SM@pixelr_222·
@nixbenoit @ChelmsDeep America has a free market economy. China, less so. So we should emulate China?
English
0
0
0
0
NixBen
NixBen@nixbenoit·
@ChelmsDeep China blocks our AI while subsidizing its own with cheap electricity. Now their models are competitive and cheaper. Real reciprocity: Ban U.S. cloud providers from hosting Chinese AI models and stop government-funded organizations from using them until China opens its market.
English
1
0
0
3
SM
SM@pixelr_222·
@neilorpen @paulmcginley_ @PolitlcsUK REPRESENTATIVE democracy. Either it represents the 0.1% or the population. Have oxbridge people, sure. Don't stuff it full of them. Or stuff it full of people from any particular background. This isn't hard Neil, you're probably just deeply elitist and classist.
English
0
0
0
1
neil orpen
neil orpen@neilorpen·
@paulmcginley_ @PolitlcsUK Ah - so it is actually your overwhelming hatred of people you see as elite It is just about your envy of an upper class
English
1
0
0
11
Politics UK
Politics UK@PolitlcsUK·
🚨 NEW: Andy Burnham has appointed the most Oxbridge-educated Cabinet in modern times 17 of the 28 Ministers appointed were educated at either Oxford or Cambridge, according to analysis by The i Paper
English
359
596
6.2K
1.6M
SM
SM@pixelr_222·
@hosseeb "The $1.5B Anthropic settlement was for pirating books. It was not for *training* on books. The judge ruled it was perfectly fine to train on copyrighted materials as long as you actually bought them." By said logic if they distilled from paid subs / API - then a-ok?
English
0
0
1
2
Haseeb >|<
Haseeb >|<@hosseeb·
Important clarification, as I keep seeing misinformation about this: The $1.5B Anthropic settlement was for pirating books. It was not for *training* on books. The judge ruled it was perfectly fine to train on copyrighted materials as long as you actually bought them. Now how does this compare to distillation? It's very unclear that distillation, which involves training on reasoning traces that you buy from a third party, is anything like this. 1) Pure AI-generated outputs are not copyrightable, according to the US Copyright Office. 2) The ToS with OAI/Ant to not train on the model traces restricts the person who signs the ToS, but it's not clear it stops them from selling their own data, or a third party from buying that data. 3) Traces are not hidden, so hard to claim traces can be treated as trade secrets. It's very unclear that distillation is illegal. It's like building a Pong knock-off game based on the original Pong. Unless Pong is patented and you violate the patent, it's normally legal and just a part of how businesses learn from each other and compete. Now, if they are creating webs of tens of thousands of accounts to distill and directly violating Anthropic's ToS, that's a stronger argument of corporate espionage. But if they're just buying this data arm's length from data brokers, good argument that it's legal to distill. (Any lawyers feel free to correct me)
0xBlackhatBull@0xBlackhatBull

@SelfRelianceOS @hosseeb Because the American models were also trained on IP Theft. See anthropics 1.5B dollar settlement that happened just this week. US labs can fuck off.

English
14
2
55
14K
SM
SM@pixelr_222·
@NovusOrderM @macastel3 Why do you think anthropic just settled for $1.5 billion The frontier labs stole their entire training material and IP. So the IP theft, repackaging and reselling. Yup, happens closer to home too.
English
0
0
0
0
Frank Ricard
Frank Ricard@NovusOrderM·
I'm all for competition. I'm all for open-source models. I am not all for Chinese IP theft, distillation, repackaging, and reselling them to US consumers and enterprises. Super unfortunate we don't have decades of evidence of industrial-scale IP theft from CCP-aligned firms to point to. x.com/NovusOrderM/st…
Frank Ricard@NovusOrderM

Well, as it relates to distillation. 1. Anthropic releases Mythos Preview in April 2026 2. Anthropic sends a letter in June 2026 to US officials accusing Alibaba of approximately 30 million exchanges with its model, using thousands of fraudulent accounts. 3. Kimi K-3 is released in July 2026 with similar frontier capabilities. Alibaba holds an approximate 36% stake in Moonshot AI. Quite the coincidence. As it relates to historic patterns of the same strategy, it's not a coincidence that China's industrial system has repeatedly combined 1) acquisition of foreign technology, coercively or illegally, 2) compression of the time and cost required for independent research and development, 3) state-supported financing, procurement, and industrial policy and capacity expansion, and 4) aggressive price competition in domestic and international markets. 2024 USTR concluded that China's tech-transfer practices persisted, that state-supported cyber theft continued, and that these practices cost American companies and consumers billions of dollars. Case examples: 1. Sinovel and American Superconductor case. A federal jury convicted China-based Sinovel of stealing American Superconductor’s wind-turbine control software. 2. Dupont titanium-dioxide technology case. Federal prosecutors proved a conspiracy to transfer DuPont’s proprietary chloride-route titanium-dioxide process to Chinese state-owned companies. 3. DuPont Pioneer and Monsanto seed lines case. A defendant connected to the Chinese agricultural company Dabeinong pleaded guilty in a conspiracy involving the theft of proprietary inbred corn seeds from DuPont Pioneer and Monsanto fields 4. Chinese intelligence targeting GE Aviation. A Chinese Ministry of State Security officer was convicted of attempting to obtain proprietary aviation information from U.S. companies, including GE Aviation. 5. Google. In January 2026, a former Google engineer was convicted of stealing thousands of pages of confidential material concerning AI supercomputing and chip infrastructure while affiliating himself with two Chinese technology ventures. Prosecutors presented evidence that he intended to reproduce or modify Google’s technology for Chinese entities. 6. Motorola Solutions / Hytera Communications. Beginning in 2006, China-based Hytera recruited Motorola engineers and directed them to take Motorola source code and other trade secrets covering digital mobile radio technology. Motorola had developed the technology through years of research and engineering. Hytera pleaded guilty to conspiracy to steal trade secrets. In March 2026, it was fined $50 million and placed on probation. 7. Coca-Cola / Weihai Jinhong Group. Chemist Xiaorong “Shannon” You obtained proprietary formulations for bisphenol-A-free—or BPA-free—coatings used inside beverage and food cans while working at Coca-Cola and Eastman Chemical. You was convicted of economic espionage, possession of stolen trade secrets and related offenses and was sentenced to 14 years in prison. 8. GlaxoSmithKline / Renopharma. Several individuals associated with GlaxoSmithKline transferred confidential pharmaceutical and biotechnology information to Renopharma, a company created to develop drugs in China. The economic value was potentially enormous: the Department of Justice noted that these kinds of products typically cost more than $1 billion each to research and develop. At least four defendants pleaded guilty, while another defendant was convicted 9. GE turbine tech / Chinese firms, universities, and government-linked industrial programs. Xiaoqing Zheng, a longtime GE Power engineer, was convicted of conspiring to steal GE trade secrets involving turbine-sealing technology and other ground-based and aviation turbine technologies. Trial evidence showed that Zheng and collaborators in China intended the information to benefit the People’s Republic of China and Chinese organizations researching, manufacturing and developing turbine components. Zheng was also identified as a participant in China’s Thousand Talents Program. Following a jury conviction, he was sentenced in 2023 to two years in prison. 10. Edwards Lifesciences / Medtronic / Chinese Medical-device company. Engineer Wenfeng Lu stole confidential information and trade secrets from two American medical-device companies that developed products used to treat cardiac and vascular conditions. While still working for the U.S. companies, Lu traveled repeatedly to China, obtained financing and prepared to open a Chinese company that would manufacture vascular-treatment devices using the stolen American technology. He applied for and was selected to receive approximately $328,000 in Chinese government funding, along with three years of free laboratory space in a Chinese technology park. Prosecutors described the program as one designed to attract overseas scientists and intellectual property to develop Chinese biomedical technology. Lu pleaded guilty to six trade-secret offenses and was sentenced to 27 months in prison. He was arrested before departing for China, preventing the proposed venture from being fully implemented. Just to name a few... Appreciate the comment, honestly. A few of these cases I was not familiar with until having AI dig into it and reading up on them. It's broader than I originally anticipated.

English
1
0
0
89
Marco Castelli
Marco Castelli@macastel3·
At the end of the day, whether a AI model is American or Chinese, who cares? Unless you are a shareholder and thus have a conflict of interest, i just want a fast easy and cheap model So i'm all for competition Also it's a race without finish line so it doesn't matter who wins today
English
32
3
64
3.6K
hypocrisy exists
hypocrisy exists@hypocrisyexists·
@AmmarKazmi We have a much lower tax rate for our low earners compared to any other European country and a higher rate for higher earners. You're talking nonsense
English
3
0
13
8.6K
SM
SM@pixelr_222·
@lomlmashiro When you surround yourself with vapid performative virtue signallers, don't be surprised when you're next to be ostracised by the dogood society.
English
0
0
0
0
Ez🪩
Ez🪩@lomlmashiro·
is it too extreme for me to cut someone off due to their excessive and unnecessary chat GPT usage?
English
321
496
16.3K
267.3K
SM
SM@pixelr_222·
@user20231399892 @randomthougtz @TheForgexx It wasn't a graphic sex scene. It's rated for 15 year olds. The sex scene in Oppenheimer set tone for the scrutiny of affair to seem intrusive. If we're to appreciate his security hearing laying bare his private life to the world, would a closed door capture that?
English
0
0
0
0
user202313213
user202313213@user20231399892·
@randomthougtz @TheForgexx i dont get why they cant just hint that such acts took place like for example, the two main characters walk off into a bedroom and shut the door behind them or somthing, why have a graphic sex scene in a movie its just fucking awkward
English
6
0
125
14.5K
The Forge 🍿
The Forge 🍿@TheForgexx·
Florence Pugh reveals the camera broke during her and Cillian Murphy’s intimate scene in Oppenheimer, leaving them stuck on set waiting for repairs. “In the middle of our sex scene, the camera broke. No one knows this, but it did.” “Our camera broke when we were both naked, and it was not ideal timing.” Pugh said the closed set made the situation awkward, with her and Murphy having to wait while the crew fixed the equipment. Instead of sitting silently, she turned the unexpected pause into a learning moment: “So tell me, what’s wrong with this camera?” Director Christopher Nolan later explained the issue involved the way light was entering the camera. Pugh said the experience showed how passionate and skilled the entire Oppenheimer crew was, describing every moment on set as a privilege.
The Forge 🍿 tweet mediaThe Forge 🍿 tweet media
English
828
1.2K
69.6K
17.9M
SM
SM@pixelr_222·
@AstUgq @19lily98 Ok baga log off the alt. 0 in 0 people recognise you after these changes.
English
0
0
0
0
Ast Ugq
Ast Ugq@AstUgq·
@19lily98 ? What's scary about that? She literally just looks thinner with better makeup. You DELIBERATELY picked a drag wig with drab makeup drab monochromatic hair where her double chin is literally enhanced and out here acting a cunt about it
English
3
0
3
1.1K
SM
SM@pixelr_222·
@chhopsky @Colin_d_m No, hand writing every line of syntax isn't the job anymore.
English
0
0
0
1
christina 死神
christina 死神@chhopsky·
@Colin_d_m “hand crafting artisanal helper functions” you’re describing programming. isn’t that.. your job
English
10
0
168
4.7K
Colin
Colin@Colin_d_m·
I hate redditors so much it's unreal. If they think anyone is still hand-crafting artisanal helper functions they're so insanely out of the loop
Colin tweet media
English
49
3
672
75.3K
SM
SM@pixelr_222·
@flyfour @kayeadams Unfair dismissal does tend to fuck your perception of an employer.
English
0
0
0
0
Flyfour
Flyfour@flyfour·
@kayeadams So the BBC has never done anything for you. Apart from employ you to present a BBC Radio Scotland show for 15 years
English
1
0
23
2K
SM
SM@pixelr_222·
@drydenwtbrown This is an argument that regulating for monopoly/oglipoly with no free market competition is a good thing because concentration of capital makes intensive R&D easier. I don't accept that position as being acceptable.
English
0
0
0
0
DRYDEN
DRYDEN@drydenwtbrown·
It's unbelievable that people are unwilling to admit that open weight models present tradeoffs regarding the speed of technological progress. Proliferation, at the cost of capex capacity of the labs that have made every single intelligence advancement so far. How does that net out? Not sure. Probably slows things down, but it's complicated. If your argument is that open weight models do not affect frontier lab revenue / margins, why do they even matter? Is this all new demand? I thought the story was enterprise switching from labs to open weight to retain their data.
DRYDEN@drydenwtbrown

It’s a simple argument. Open-weight models compress frontier model margins, which reduces cash to invest in scaling up compute for larger pretrains — which is where material increases in intelligence come from. Why feign incomprehension?

English
14
0
32
9.9K